xref: /linux/kernel/fork.c (revision 3b64b1881143ce9e461c211cc81acc72d0cdc476)
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6 
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13 
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/fs.h>
31 #include <linux/nsproxy.h>
32 #include <linux/capability.h>
33 #include <linux/cpu.h>
34 #include <linux/cgroup.h>
35 #include <linux/security.h>
36 #include <linux/hugetlb.h>
37 #include <linux/seccomp.h>
38 #include <linux/swap.h>
39 #include <linux/syscalls.h>
40 #include <linux/jiffies.h>
41 #include <linux/futex.h>
42 #include <linux/compat.h>
43 #include <linux/kthread.h>
44 #include <linux/task_io_accounting_ops.h>
45 #include <linux/rcupdate.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/audit.h>
49 #include <linux/memcontrol.h>
50 #include <linux/ftrace.h>
51 #include <linux/proc_fs.h>
52 #include <linux/profile.h>
53 #include <linux/rmap.h>
54 #include <linux/ksm.h>
55 #include <linux/acct.h>
56 #include <linux/tsacct_kern.h>
57 #include <linux/cn_proc.h>
58 #include <linux/freezer.h>
59 #include <linux/delayacct.h>
60 #include <linux/taskstats_kern.h>
61 #include <linux/random.h>
62 #include <linux/tty.h>
63 #include <linux/blkdev.h>
64 #include <linux/fs_struct.h>
65 #include <linux/magic.h>
66 #include <linux/perf_event.h>
67 #include <linux/posix-timers.h>
68 #include <linux/user-return-notifier.h>
69 #include <linux/oom.h>
70 #include <linux/khugepaged.h>
71 #include <linux/signalfd.h>
72 #include <linux/uprobes.h>
73 
74 #include <asm/pgtable.h>
75 #include <asm/pgalloc.h>
76 #include <asm/uaccess.h>
77 #include <asm/mmu_context.h>
78 #include <asm/cacheflush.h>
79 #include <asm/tlbflush.h>
80 
81 #include <trace/events/sched.h>
82 
83 #define CREATE_TRACE_POINTS
84 #include <trace/events/task.h>
85 
86 /*
87  * Protected counters by write_lock_irq(&tasklist_lock)
88  */
89 unsigned long total_forks;	/* Handle normal Linux uptimes. */
90 int nr_threads;			/* The idle threads do not count.. */
91 
92 int max_threads;		/* tunable limit on nr_threads */
93 
94 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
95 
96 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
97 
98 #ifdef CONFIG_PROVE_RCU
99 int lockdep_tasklist_lock_is_held(void)
100 {
101 	return lockdep_is_held(&tasklist_lock);
102 }
103 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
104 #endif /* #ifdef CONFIG_PROVE_RCU */
105 
106 int nr_processes(void)
107 {
108 	int cpu;
109 	int total = 0;
110 
111 	for_each_possible_cpu(cpu)
112 		total += per_cpu(process_counts, cpu);
113 
114 	return total;
115 }
116 
117 void __weak arch_release_task_struct(struct task_struct *tsk)
118 {
119 }
120 
121 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
122 static struct kmem_cache *task_struct_cachep;
123 
124 static inline struct task_struct *alloc_task_struct_node(int node)
125 {
126 	return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
127 }
128 
129 static inline void free_task_struct(struct task_struct *tsk)
130 {
131 	kmem_cache_free(task_struct_cachep, tsk);
132 }
133 #endif
134 
135 void __weak arch_release_thread_info(struct thread_info *ti)
136 {
137 }
138 
139 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
140 
141 /*
142  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
143  * kmemcache based allocator.
144  */
145 # if THREAD_SIZE >= PAGE_SIZE
146 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
147 						  int node)
148 {
149 	struct page *page = alloc_pages_node(node, THREADINFO_GFP,
150 					     THREAD_SIZE_ORDER);
151 
152 	return page ? page_address(page) : NULL;
153 }
154 
155 static inline void free_thread_info(struct thread_info *ti)
156 {
157 	free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
158 }
159 # else
160 static struct kmem_cache *thread_info_cache;
161 
162 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
163 						  int node)
164 {
165 	return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
166 }
167 
168 static void free_thread_info(struct thread_info *ti)
169 {
170 	kmem_cache_free(thread_info_cache, ti);
171 }
172 
173 void thread_info_cache_init(void)
174 {
175 	thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
176 					      THREAD_SIZE, 0, NULL);
177 	BUG_ON(thread_info_cache == NULL);
178 }
179 # endif
180 #endif
181 
182 /* SLAB cache for signal_struct structures (tsk->signal) */
183 static struct kmem_cache *signal_cachep;
184 
185 /* SLAB cache for sighand_struct structures (tsk->sighand) */
186 struct kmem_cache *sighand_cachep;
187 
188 /* SLAB cache for files_struct structures (tsk->files) */
189 struct kmem_cache *files_cachep;
190 
191 /* SLAB cache for fs_struct structures (tsk->fs) */
192 struct kmem_cache *fs_cachep;
193 
194 /* SLAB cache for vm_area_struct structures */
195 struct kmem_cache *vm_area_cachep;
196 
197 /* SLAB cache for mm_struct structures (tsk->mm) */
198 static struct kmem_cache *mm_cachep;
199 
200 static void account_kernel_stack(struct thread_info *ti, int account)
201 {
202 	struct zone *zone = page_zone(virt_to_page(ti));
203 
204 	mod_zone_page_state(zone, NR_KERNEL_STACK, account);
205 }
206 
207 void free_task(struct task_struct *tsk)
208 {
209 	account_kernel_stack(tsk->stack, -1);
210 	arch_release_thread_info(tsk->stack);
211 	free_thread_info(tsk->stack);
212 	rt_mutex_debug_task_free(tsk);
213 	ftrace_graph_exit_task(tsk);
214 	put_seccomp_filter(tsk);
215 	arch_release_task_struct(tsk);
216 	free_task_struct(tsk);
217 }
218 EXPORT_SYMBOL(free_task);
219 
220 static inline void free_signal_struct(struct signal_struct *sig)
221 {
222 	taskstats_tgid_free(sig);
223 	sched_autogroup_exit(sig);
224 	kmem_cache_free(signal_cachep, sig);
225 }
226 
227 static inline void put_signal_struct(struct signal_struct *sig)
228 {
229 	if (atomic_dec_and_test(&sig->sigcnt))
230 		free_signal_struct(sig);
231 }
232 
233 void __put_task_struct(struct task_struct *tsk)
234 {
235 	WARN_ON(!tsk->exit_state);
236 	WARN_ON(atomic_read(&tsk->usage));
237 	WARN_ON(tsk == current);
238 
239 	security_task_free(tsk);
240 	exit_creds(tsk);
241 	delayacct_tsk_free(tsk);
242 	put_signal_struct(tsk->signal);
243 
244 	if (!profile_handoff_task(tsk))
245 		free_task(tsk);
246 }
247 EXPORT_SYMBOL_GPL(__put_task_struct);
248 
249 void __init __weak arch_task_cache_init(void) { }
250 
251 void __init fork_init(unsigned long mempages)
252 {
253 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
254 #ifndef ARCH_MIN_TASKALIGN
255 #define ARCH_MIN_TASKALIGN	L1_CACHE_BYTES
256 #endif
257 	/* create a slab on which task_structs can be allocated */
258 	task_struct_cachep =
259 		kmem_cache_create("task_struct", sizeof(struct task_struct),
260 			ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
261 #endif
262 
263 	/* do the arch specific task caches init */
264 	arch_task_cache_init();
265 
266 	/*
267 	 * The default maximum number of threads is set to a safe
268 	 * value: the thread structures can take up at most half
269 	 * of memory.
270 	 */
271 	max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
272 
273 	/*
274 	 * we need to allow at least 20 threads to boot a system
275 	 */
276 	if (max_threads < 20)
277 		max_threads = 20;
278 
279 	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
280 	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
281 	init_task.signal->rlim[RLIMIT_SIGPENDING] =
282 		init_task.signal->rlim[RLIMIT_NPROC];
283 }
284 
285 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
286 					       struct task_struct *src)
287 {
288 	*dst = *src;
289 	return 0;
290 }
291 
292 static struct task_struct *dup_task_struct(struct task_struct *orig)
293 {
294 	struct task_struct *tsk;
295 	struct thread_info *ti;
296 	unsigned long *stackend;
297 	int node = tsk_fork_get_node(orig);
298 	int err;
299 
300 	tsk = alloc_task_struct_node(node);
301 	if (!tsk)
302 		return NULL;
303 
304 	ti = alloc_thread_info_node(tsk, node);
305 	if (!ti)
306 		goto free_tsk;
307 
308 	err = arch_dup_task_struct(tsk, orig);
309 	if (err)
310 		goto free_ti;
311 
312 	tsk->stack = ti;
313 
314 	setup_thread_stack(tsk, orig);
315 	clear_user_return_notifier(tsk);
316 	clear_tsk_need_resched(tsk);
317 	stackend = end_of_stack(tsk);
318 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
319 
320 #ifdef CONFIG_CC_STACKPROTECTOR
321 	tsk->stack_canary = get_random_int();
322 #endif
323 
324 	/*
325 	 * One for us, one for whoever does the "release_task()" (usually
326 	 * parent)
327 	 */
328 	atomic_set(&tsk->usage, 2);
329 #ifdef CONFIG_BLK_DEV_IO_TRACE
330 	tsk->btrace_seq = 0;
331 #endif
332 	tsk->splice_pipe = NULL;
333 	tsk->task_frag.page = NULL;
334 
335 	account_kernel_stack(ti, 1);
336 
337 	return tsk;
338 
339 free_ti:
340 	free_thread_info(ti);
341 free_tsk:
342 	free_task_struct(tsk);
343 	return NULL;
344 }
345 
346 #ifdef CONFIG_MMU
347 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
348 {
349 	struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
350 	struct rb_node **rb_link, *rb_parent;
351 	int retval;
352 	unsigned long charge;
353 	struct mempolicy *pol;
354 
355 	down_write(&oldmm->mmap_sem);
356 	flush_cache_dup_mm(oldmm);
357 	uprobe_dup_mmap(oldmm, mm);
358 	/*
359 	 * Not linked in yet - no deadlock potential:
360 	 */
361 	down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
362 
363 	mm->locked_vm = 0;
364 	mm->mmap = NULL;
365 	mm->mmap_cache = NULL;
366 	mm->free_area_cache = oldmm->mmap_base;
367 	mm->cached_hole_size = ~0UL;
368 	mm->map_count = 0;
369 	cpumask_clear(mm_cpumask(mm));
370 	mm->mm_rb = RB_ROOT;
371 	rb_link = &mm->mm_rb.rb_node;
372 	rb_parent = NULL;
373 	pprev = &mm->mmap;
374 	retval = ksm_fork(mm, oldmm);
375 	if (retval)
376 		goto out;
377 	retval = khugepaged_fork(mm, oldmm);
378 	if (retval)
379 		goto out;
380 
381 	prev = NULL;
382 	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
383 		struct file *file;
384 
385 		if (mpnt->vm_flags & VM_DONTCOPY) {
386 			vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
387 							-vma_pages(mpnt));
388 			continue;
389 		}
390 		charge = 0;
391 		if (mpnt->vm_flags & VM_ACCOUNT) {
392 			unsigned long len = vma_pages(mpnt);
393 
394 			if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
395 				goto fail_nomem;
396 			charge = len;
397 		}
398 		tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
399 		if (!tmp)
400 			goto fail_nomem;
401 		*tmp = *mpnt;
402 		INIT_LIST_HEAD(&tmp->anon_vma_chain);
403 		pol = mpol_dup(vma_policy(mpnt));
404 		retval = PTR_ERR(pol);
405 		if (IS_ERR(pol))
406 			goto fail_nomem_policy;
407 		vma_set_policy(tmp, pol);
408 		tmp->vm_mm = mm;
409 		if (anon_vma_fork(tmp, mpnt))
410 			goto fail_nomem_anon_vma_fork;
411 		tmp->vm_flags &= ~VM_LOCKED;
412 		tmp->vm_next = tmp->vm_prev = NULL;
413 		file = tmp->vm_file;
414 		if (file) {
415 			struct inode *inode = file->f_path.dentry->d_inode;
416 			struct address_space *mapping = file->f_mapping;
417 
418 			get_file(file);
419 			if (tmp->vm_flags & VM_DENYWRITE)
420 				atomic_dec(&inode->i_writecount);
421 			mutex_lock(&mapping->i_mmap_mutex);
422 			if (tmp->vm_flags & VM_SHARED)
423 				mapping->i_mmap_writable++;
424 			flush_dcache_mmap_lock(mapping);
425 			/* insert tmp into the share list, just after mpnt */
426 			vma_prio_tree_add(tmp, mpnt);
427 			flush_dcache_mmap_unlock(mapping);
428 			mutex_unlock(&mapping->i_mmap_mutex);
429 		}
430 
431 		/*
432 		 * Clear hugetlb-related page reserves for children. This only
433 		 * affects MAP_PRIVATE mappings. Faults generated by the child
434 		 * are not guaranteed to succeed, even if read-only
435 		 */
436 		if (is_vm_hugetlb_page(tmp))
437 			reset_vma_resv_huge_pages(tmp);
438 
439 		/*
440 		 * Link in the new vma and copy the page table entries.
441 		 */
442 		*pprev = tmp;
443 		pprev = &tmp->vm_next;
444 		tmp->vm_prev = prev;
445 		prev = tmp;
446 
447 		__vma_link_rb(mm, tmp, rb_link, rb_parent);
448 		rb_link = &tmp->vm_rb.rb_right;
449 		rb_parent = &tmp->vm_rb;
450 
451 		mm->map_count++;
452 		retval = copy_page_range(mm, oldmm, mpnt);
453 
454 		if (tmp->vm_ops && tmp->vm_ops->open)
455 			tmp->vm_ops->open(tmp);
456 
457 		if (retval)
458 			goto out;
459 	}
460 	/* a new mm has just been created */
461 	arch_dup_mmap(oldmm, mm);
462 	retval = 0;
463 out:
464 	up_write(&mm->mmap_sem);
465 	flush_tlb_mm(oldmm);
466 	up_write(&oldmm->mmap_sem);
467 	return retval;
468 fail_nomem_anon_vma_fork:
469 	mpol_put(pol);
470 fail_nomem_policy:
471 	kmem_cache_free(vm_area_cachep, tmp);
472 fail_nomem:
473 	retval = -ENOMEM;
474 	vm_unacct_memory(charge);
475 	goto out;
476 }
477 
478 static inline int mm_alloc_pgd(struct mm_struct *mm)
479 {
480 	mm->pgd = pgd_alloc(mm);
481 	if (unlikely(!mm->pgd))
482 		return -ENOMEM;
483 	return 0;
484 }
485 
486 static inline void mm_free_pgd(struct mm_struct *mm)
487 {
488 	pgd_free(mm, mm->pgd);
489 }
490 #else
491 #define dup_mmap(mm, oldmm)	(0)
492 #define mm_alloc_pgd(mm)	(0)
493 #define mm_free_pgd(mm)
494 #endif /* CONFIG_MMU */
495 
496 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
497 
498 #define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
499 #define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
500 
501 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
502 
503 static int __init coredump_filter_setup(char *s)
504 {
505 	default_dump_filter =
506 		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
507 		MMF_DUMP_FILTER_MASK;
508 	return 1;
509 }
510 
511 __setup("coredump_filter=", coredump_filter_setup);
512 
513 #include <linux/init_task.h>
514 
515 static void mm_init_aio(struct mm_struct *mm)
516 {
517 #ifdef CONFIG_AIO
518 	spin_lock_init(&mm->ioctx_lock);
519 	INIT_HLIST_HEAD(&mm->ioctx_list);
520 #endif
521 }
522 
523 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
524 {
525 	atomic_set(&mm->mm_users, 1);
526 	atomic_set(&mm->mm_count, 1);
527 	init_rwsem(&mm->mmap_sem);
528 	INIT_LIST_HEAD(&mm->mmlist);
529 	mm->flags = (current->mm) ?
530 		(current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
531 	mm->core_state = NULL;
532 	mm->nr_ptes = 0;
533 	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
534 	spin_lock_init(&mm->page_table_lock);
535 	mm->free_area_cache = TASK_UNMAPPED_BASE;
536 	mm->cached_hole_size = ~0UL;
537 	mm_init_aio(mm);
538 	mm_init_owner(mm, p);
539 
540 	if (likely(!mm_alloc_pgd(mm))) {
541 		mm->def_flags = 0;
542 		mmu_notifier_mm_init(mm);
543 		return mm;
544 	}
545 
546 	free_mm(mm);
547 	return NULL;
548 }
549 
550 static void check_mm(struct mm_struct *mm)
551 {
552 	int i;
553 
554 	for (i = 0; i < NR_MM_COUNTERS; i++) {
555 		long x = atomic_long_read(&mm->rss_stat.count[i]);
556 
557 		if (unlikely(x))
558 			printk(KERN_ALERT "BUG: Bad rss-counter state "
559 					  "mm:%p idx:%d val:%ld\n", mm, i, x);
560 	}
561 
562 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
563 	VM_BUG_ON(mm->pmd_huge_pte);
564 #endif
565 }
566 
567 /*
568  * Allocate and initialize an mm_struct.
569  */
570 struct mm_struct *mm_alloc(void)
571 {
572 	struct mm_struct *mm;
573 
574 	mm = allocate_mm();
575 	if (!mm)
576 		return NULL;
577 
578 	memset(mm, 0, sizeof(*mm));
579 	mm_init_cpumask(mm);
580 	return mm_init(mm, current);
581 }
582 
583 /*
584  * Called when the last reference to the mm
585  * is dropped: either by a lazy thread or by
586  * mmput. Free the page directory and the mm.
587  */
588 void __mmdrop(struct mm_struct *mm)
589 {
590 	BUG_ON(mm == &init_mm);
591 	mm_free_pgd(mm);
592 	destroy_context(mm);
593 	mmu_notifier_mm_destroy(mm);
594 	check_mm(mm);
595 	free_mm(mm);
596 }
597 EXPORT_SYMBOL_GPL(__mmdrop);
598 
599 /*
600  * Decrement the use count and release all resources for an mm.
601  */
602 void mmput(struct mm_struct *mm)
603 {
604 	might_sleep();
605 
606 	if (atomic_dec_and_test(&mm->mm_users)) {
607 		uprobe_clear_state(mm);
608 		exit_aio(mm);
609 		ksm_exit(mm);
610 		khugepaged_exit(mm); /* must run before exit_mmap */
611 		exit_mmap(mm);
612 		set_mm_exe_file(mm, NULL);
613 		if (!list_empty(&mm->mmlist)) {
614 			spin_lock(&mmlist_lock);
615 			list_del(&mm->mmlist);
616 			spin_unlock(&mmlist_lock);
617 		}
618 		if (mm->binfmt)
619 			module_put(mm->binfmt->module);
620 		mmdrop(mm);
621 	}
622 }
623 EXPORT_SYMBOL_GPL(mmput);
624 
625 /*
626  * We added or removed a vma mapping the executable. The vmas are only mapped
627  * during exec and are not mapped with the mmap system call.
628  * Callers must hold down_write() on the mm's mmap_sem for these
629  */
630 void added_exe_file_vma(struct mm_struct *mm)
631 {
632 	mm->num_exe_file_vmas++;
633 }
634 
635 void removed_exe_file_vma(struct mm_struct *mm)
636 {
637 	mm->num_exe_file_vmas--;
638 	if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
639 		fput(mm->exe_file);
640 		mm->exe_file = NULL;
641 	}
642 
643 }
644 
645 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
646 {
647 	if (new_exe_file)
648 		get_file(new_exe_file);
649 	if (mm->exe_file)
650 		fput(mm->exe_file);
651 	mm->exe_file = new_exe_file;
652 	mm->num_exe_file_vmas = 0;
653 }
654 
655 struct file *get_mm_exe_file(struct mm_struct *mm)
656 {
657 	struct file *exe_file;
658 
659 	/* We need mmap_sem to protect against races with removal of
660 	 * VM_EXECUTABLE vmas */
661 	down_read(&mm->mmap_sem);
662 	exe_file = mm->exe_file;
663 	if (exe_file)
664 		get_file(exe_file);
665 	up_read(&mm->mmap_sem);
666 	return exe_file;
667 }
668 
669 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
670 {
671 	/* It's safe to write the exe_file pointer without exe_file_lock because
672 	 * this is called during fork when the task is not yet in /proc */
673 	newmm->exe_file = get_mm_exe_file(oldmm);
674 }
675 
676 /**
677  * get_task_mm - acquire a reference to the task's mm
678  *
679  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
680  * this kernel workthread has transiently adopted a user mm with use_mm,
681  * to do its AIO) is not set and if so returns a reference to it, after
682  * bumping up the use count.  User must release the mm via mmput()
683  * after use.  Typically used by /proc and ptrace.
684  */
685 struct mm_struct *get_task_mm(struct task_struct *task)
686 {
687 	struct mm_struct *mm;
688 
689 	task_lock(task);
690 	mm = task->mm;
691 	if (mm) {
692 		if (task->flags & PF_KTHREAD)
693 			mm = NULL;
694 		else
695 			atomic_inc(&mm->mm_users);
696 	}
697 	task_unlock(task);
698 	return mm;
699 }
700 EXPORT_SYMBOL_GPL(get_task_mm);
701 
702 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
703 {
704 	struct mm_struct *mm;
705 	int err;
706 
707 	err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
708 	if (err)
709 		return ERR_PTR(err);
710 
711 	mm = get_task_mm(task);
712 	if (mm && mm != current->mm &&
713 			!ptrace_may_access(task, mode)) {
714 		mmput(mm);
715 		mm = ERR_PTR(-EACCES);
716 	}
717 	mutex_unlock(&task->signal->cred_guard_mutex);
718 
719 	return mm;
720 }
721 
722 static void complete_vfork_done(struct task_struct *tsk)
723 {
724 	struct completion *vfork;
725 
726 	task_lock(tsk);
727 	vfork = tsk->vfork_done;
728 	if (likely(vfork)) {
729 		tsk->vfork_done = NULL;
730 		complete(vfork);
731 	}
732 	task_unlock(tsk);
733 }
734 
735 static int wait_for_vfork_done(struct task_struct *child,
736 				struct completion *vfork)
737 {
738 	int killed;
739 
740 	freezer_do_not_count();
741 	killed = wait_for_completion_killable(vfork);
742 	freezer_count();
743 
744 	if (killed) {
745 		task_lock(child);
746 		child->vfork_done = NULL;
747 		task_unlock(child);
748 	}
749 
750 	put_task_struct(child);
751 	return killed;
752 }
753 
754 /* Please note the differences between mmput and mm_release.
755  * mmput is called whenever we stop holding onto a mm_struct,
756  * error success whatever.
757  *
758  * mm_release is called after a mm_struct has been removed
759  * from the current process.
760  *
761  * This difference is important for error handling, when we
762  * only half set up a mm_struct for a new process and need to restore
763  * the old one.  Because we mmput the new mm_struct before
764  * restoring the old one. . .
765  * Eric Biederman 10 January 1998
766  */
767 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
768 {
769 	/* Get rid of any futexes when releasing the mm */
770 #ifdef CONFIG_FUTEX
771 	if (unlikely(tsk->robust_list)) {
772 		exit_robust_list(tsk);
773 		tsk->robust_list = NULL;
774 	}
775 #ifdef CONFIG_COMPAT
776 	if (unlikely(tsk->compat_robust_list)) {
777 		compat_exit_robust_list(tsk);
778 		tsk->compat_robust_list = NULL;
779 	}
780 #endif
781 	if (unlikely(!list_empty(&tsk->pi_state_list)))
782 		exit_pi_state_list(tsk);
783 #endif
784 
785 	uprobe_free_utask(tsk);
786 
787 	/* Get rid of any cached register state */
788 	deactivate_mm(tsk, mm);
789 
790 	/*
791 	 * If we're exiting normally, clear a user-space tid field if
792 	 * requested.  We leave this alone when dying by signal, to leave
793 	 * the value intact in a core dump, and to save the unnecessary
794 	 * trouble, say, a killed vfork parent shouldn't touch this mm.
795 	 * Userland only wants this done for a sys_exit.
796 	 */
797 	if (tsk->clear_child_tid) {
798 		if (!(tsk->flags & PF_SIGNALED) &&
799 		    atomic_read(&mm->mm_users) > 1) {
800 			/*
801 			 * We don't check the error code - if userspace has
802 			 * not set up a proper pointer then tough luck.
803 			 */
804 			put_user(0, tsk->clear_child_tid);
805 			sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
806 					1, NULL, NULL, 0);
807 		}
808 		tsk->clear_child_tid = NULL;
809 	}
810 
811 	/*
812 	 * All done, finally we can wake up parent and return this mm to him.
813 	 * Also kthread_stop() uses this completion for synchronization.
814 	 */
815 	if (tsk->vfork_done)
816 		complete_vfork_done(tsk);
817 }
818 
819 /*
820  * Allocate a new mm structure and copy contents from the
821  * mm structure of the passed in task structure.
822  */
823 struct mm_struct *dup_mm(struct task_struct *tsk)
824 {
825 	struct mm_struct *mm, *oldmm = current->mm;
826 	int err;
827 
828 	if (!oldmm)
829 		return NULL;
830 
831 	mm = allocate_mm();
832 	if (!mm)
833 		goto fail_nomem;
834 
835 	memcpy(mm, oldmm, sizeof(*mm));
836 	mm_init_cpumask(mm);
837 
838 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
839 	mm->pmd_huge_pte = NULL;
840 #endif
841 	if (!mm_init(mm, tsk))
842 		goto fail_nomem;
843 
844 	if (init_new_context(tsk, mm))
845 		goto fail_nocontext;
846 
847 	dup_mm_exe_file(oldmm, mm);
848 
849 	err = dup_mmap(mm, oldmm);
850 	if (err)
851 		goto free_pt;
852 
853 	mm->hiwater_rss = get_mm_rss(mm);
854 	mm->hiwater_vm = mm->total_vm;
855 
856 	if (mm->binfmt && !try_module_get(mm->binfmt->module))
857 		goto free_pt;
858 
859 	return mm;
860 
861 free_pt:
862 	/* don't put binfmt in mmput, we haven't got module yet */
863 	mm->binfmt = NULL;
864 	mmput(mm);
865 
866 fail_nomem:
867 	return NULL;
868 
869 fail_nocontext:
870 	/*
871 	 * If init_new_context() failed, we cannot use mmput() to free the mm
872 	 * because it calls destroy_context()
873 	 */
874 	mm_free_pgd(mm);
875 	free_mm(mm);
876 	return NULL;
877 }
878 
879 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
880 {
881 	struct mm_struct *mm, *oldmm;
882 	int retval;
883 
884 	tsk->min_flt = tsk->maj_flt = 0;
885 	tsk->nvcsw = tsk->nivcsw = 0;
886 #ifdef CONFIG_DETECT_HUNG_TASK
887 	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
888 #endif
889 
890 	tsk->mm = NULL;
891 	tsk->active_mm = NULL;
892 
893 	/*
894 	 * Are we cloning a kernel thread?
895 	 *
896 	 * We need to steal a active VM for that..
897 	 */
898 	oldmm = current->mm;
899 	if (!oldmm)
900 		return 0;
901 
902 	if (clone_flags & CLONE_VM) {
903 		atomic_inc(&oldmm->mm_users);
904 		mm = oldmm;
905 		goto good_mm;
906 	}
907 
908 	retval = -ENOMEM;
909 	mm = dup_mm(tsk);
910 	if (!mm)
911 		goto fail_nomem;
912 
913 good_mm:
914 	tsk->mm = mm;
915 	tsk->active_mm = mm;
916 	return 0;
917 
918 fail_nomem:
919 	return retval;
920 }
921 
922 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
923 {
924 	struct fs_struct *fs = current->fs;
925 	if (clone_flags & CLONE_FS) {
926 		/* tsk->fs is already what we want */
927 		spin_lock(&fs->lock);
928 		if (fs->in_exec) {
929 			spin_unlock(&fs->lock);
930 			return -EAGAIN;
931 		}
932 		fs->users++;
933 		spin_unlock(&fs->lock);
934 		return 0;
935 	}
936 	tsk->fs = copy_fs_struct(fs);
937 	if (!tsk->fs)
938 		return -ENOMEM;
939 	return 0;
940 }
941 
942 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
943 {
944 	struct files_struct *oldf, *newf;
945 	int error = 0;
946 
947 	/*
948 	 * A background process may not have any files ...
949 	 */
950 	oldf = current->files;
951 	if (!oldf)
952 		goto out;
953 
954 	if (clone_flags & CLONE_FILES) {
955 		atomic_inc(&oldf->count);
956 		goto out;
957 	}
958 
959 	newf = dup_fd(oldf, &error);
960 	if (!newf)
961 		goto out;
962 
963 	tsk->files = newf;
964 	error = 0;
965 out:
966 	return error;
967 }
968 
969 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
970 {
971 #ifdef CONFIG_BLOCK
972 	struct io_context *ioc = current->io_context;
973 	struct io_context *new_ioc;
974 
975 	if (!ioc)
976 		return 0;
977 	/*
978 	 * Share io context with parent, if CLONE_IO is set
979 	 */
980 	if (clone_flags & CLONE_IO) {
981 		ioc_task_link(ioc);
982 		tsk->io_context = ioc;
983 	} else if (ioprio_valid(ioc->ioprio)) {
984 		new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
985 		if (unlikely(!new_ioc))
986 			return -ENOMEM;
987 
988 		new_ioc->ioprio = ioc->ioprio;
989 		put_io_context(new_ioc);
990 	}
991 #endif
992 	return 0;
993 }
994 
995 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
996 {
997 	struct sighand_struct *sig;
998 
999 	if (clone_flags & CLONE_SIGHAND) {
1000 		atomic_inc(&current->sighand->count);
1001 		return 0;
1002 	}
1003 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1004 	rcu_assign_pointer(tsk->sighand, sig);
1005 	if (!sig)
1006 		return -ENOMEM;
1007 	atomic_set(&sig->count, 1);
1008 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1009 	return 0;
1010 }
1011 
1012 void __cleanup_sighand(struct sighand_struct *sighand)
1013 {
1014 	if (atomic_dec_and_test(&sighand->count)) {
1015 		signalfd_cleanup(sighand);
1016 		kmem_cache_free(sighand_cachep, sighand);
1017 	}
1018 }
1019 
1020 
1021 /*
1022  * Initialize POSIX timer handling for a thread group.
1023  */
1024 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1025 {
1026 	unsigned long cpu_limit;
1027 
1028 	/* Thread group counters. */
1029 	thread_group_cputime_init(sig);
1030 
1031 	cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1032 	if (cpu_limit != RLIM_INFINITY) {
1033 		sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1034 		sig->cputimer.running = 1;
1035 	}
1036 
1037 	/* The timer lists. */
1038 	INIT_LIST_HEAD(&sig->cpu_timers[0]);
1039 	INIT_LIST_HEAD(&sig->cpu_timers[1]);
1040 	INIT_LIST_HEAD(&sig->cpu_timers[2]);
1041 }
1042 
1043 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1044 {
1045 	struct signal_struct *sig;
1046 
1047 	if (clone_flags & CLONE_THREAD)
1048 		return 0;
1049 
1050 	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1051 	tsk->signal = sig;
1052 	if (!sig)
1053 		return -ENOMEM;
1054 
1055 	sig->nr_threads = 1;
1056 	atomic_set(&sig->live, 1);
1057 	atomic_set(&sig->sigcnt, 1);
1058 	init_waitqueue_head(&sig->wait_chldexit);
1059 	if (clone_flags & CLONE_NEWPID)
1060 		sig->flags |= SIGNAL_UNKILLABLE;
1061 	sig->curr_target = tsk;
1062 	init_sigpending(&sig->shared_pending);
1063 	INIT_LIST_HEAD(&sig->posix_timers);
1064 
1065 	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1066 	sig->real_timer.function = it_real_fn;
1067 
1068 	task_lock(current->group_leader);
1069 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1070 	task_unlock(current->group_leader);
1071 
1072 	posix_cpu_timers_init_group(sig);
1073 
1074 	tty_audit_fork(sig);
1075 	sched_autogroup_fork(sig);
1076 
1077 #ifdef CONFIG_CGROUPS
1078 	init_rwsem(&sig->group_rwsem);
1079 #endif
1080 
1081 	sig->oom_adj = current->signal->oom_adj;
1082 	sig->oom_score_adj = current->signal->oom_score_adj;
1083 	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1084 
1085 	sig->has_child_subreaper = current->signal->has_child_subreaper ||
1086 				   current->signal->is_child_subreaper;
1087 
1088 	mutex_init(&sig->cred_guard_mutex);
1089 
1090 	return 0;
1091 }
1092 
1093 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1094 {
1095 	unsigned long new_flags = p->flags;
1096 
1097 	new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1098 	new_flags |= PF_FORKNOEXEC;
1099 	p->flags = new_flags;
1100 }
1101 
1102 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1103 {
1104 	current->clear_child_tid = tidptr;
1105 
1106 	return task_pid_vnr(current);
1107 }
1108 
1109 static void rt_mutex_init_task(struct task_struct *p)
1110 {
1111 	raw_spin_lock_init(&p->pi_lock);
1112 #ifdef CONFIG_RT_MUTEXES
1113 	plist_head_init(&p->pi_waiters);
1114 	p->pi_blocked_on = NULL;
1115 #endif
1116 }
1117 
1118 #ifdef CONFIG_MM_OWNER
1119 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1120 {
1121 	mm->owner = p;
1122 }
1123 #endif /* CONFIG_MM_OWNER */
1124 
1125 /*
1126  * Initialize POSIX timer handling for a single task.
1127  */
1128 static void posix_cpu_timers_init(struct task_struct *tsk)
1129 {
1130 	tsk->cputime_expires.prof_exp = 0;
1131 	tsk->cputime_expires.virt_exp = 0;
1132 	tsk->cputime_expires.sched_exp = 0;
1133 	INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1134 	INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1135 	INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1136 }
1137 
1138 /*
1139  * This creates a new process as a copy of the old one,
1140  * but does not actually start it yet.
1141  *
1142  * It copies the registers, and all the appropriate
1143  * parts of the process environment (as per the clone
1144  * flags). The actual kick-off is left to the caller.
1145  */
1146 static struct task_struct *copy_process(unsigned long clone_flags,
1147 					unsigned long stack_start,
1148 					struct pt_regs *regs,
1149 					unsigned long stack_size,
1150 					int __user *child_tidptr,
1151 					struct pid *pid,
1152 					int trace)
1153 {
1154 	int retval;
1155 	struct task_struct *p;
1156 	int cgroup_callbacks_done = 0;
1157 
1158 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1159 		return ERR_PTR(-EINVAL);
1160 
1161 	/*
1162 	 * Thread groups must share signals as well, and detached threads
1163 	 * can only be started up within the thread group.
1164 	 */
1165 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1166 		return ERR_PTR(-EINVAL);
1167 
1168 	/*
1169 	 * Shared signal handlers imply shared VM. By way of the above,
1170 	 * thread groups also imply shared VM. Blocking this case allows
1171 	 * for various simplifications in other code.
1172 	 */
1173 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1174 		return ERR_PTR(-EINVAL);
1175 
1176 	/*
1177 	 * Siblings of global init remain as zombies on exit since they are
1178 	 * not reaped by their parent (swapper). To solve this and to avoid
1179 	 * multi-rooted process trees, prevent global and container-inits
1180 	 * from creating siblings.
1181 	 */
1182 	if ((clone_flags & CLONE_PARENT) &&
1183 				current->signal->flags & SIGNAL_UNKILLABLE)
1184 		return ERR_PTR(-EINVAL);
1185 
1186 	retval = security_task_create(clone_flags);
1187 	if (retval)
1188 		goto fork_out;
1189 
1190 	retval = -ENOMEM;
1191 	p = dup_task_struct(current);
1192 	if (!p)
1193 		goto fork_out;
1194 
1195 	ftrace_graph_init_task(p);
1196 	get_seccomp_filter(p);
1197 
1198 	rt_mutex_init_task(p);
1199 
1200 #ifdef CONFIG_PROVE_LOCKING
1201 	DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1202 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1203 #endif
1204 	retval = -EAGAIN;
1205 	if (atomic_read(&p->real_cred->user->processes) >=
1206 			task_rlimit(p, RLIMIT_NPROC)) {
1207 		if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1208 		    p->real_cred->user != INIT_USER)
1209 			goto bad_fork_free;
1210 	}
1211 	current->flags &= ~PF_NPROC_EXCEEDED;
1212 
1213 	retval = copy_creds(p, clone_flags);
1214 	if (retval < 0)
1215 		goto bad_fork_free;
1216 
1217 	/*
1218 	 * If multiple threads are within copy_process(), then this check
1219 	 * triggers too late. This doesn't hurt, the check is only there
1220 	 * to stop root fork bombs.
1221 	 */
1222 	retval = -EAGAIN;
1223 	if (nr_threads >= max_threads)
1224 		goto bad_fork_cleanup_count;
1225 
1226 	if (!try_module_get(task_thread_info(p)->exec_domain->module))
1227 		goto bad_fork_cleanup_count;
1228 
1229 	p->did_exec = 0;
1230 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
1231 	copy_flags(clone_flags, p);
1232 	INIT_LIST_HEAD(&p->children);
1233 	INIT_LIST_HEAD(&p->sibling);
1234 	rcu_copy_process(p);
1235 	p->vfork_done = NULL;
1236 	spin_lock_init(&p->alloc_lock);
1237 
1238 	init_sigpending(&p->pending);
1239 
1240 	p->utime = p->stime = p->gtime = 0;
1241 	p->utimescaled = p->stimescaled = 0;
1242 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1243 	p->prev_utime = p->prev_stime = 0;
1244 #endif
1245 #if defined(SPLIT_RSS_COUNTING)
1246 	memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1247 #endif
1248 
1249 	p->default_timer_slack_ns = current->timer_slack_ns;
1250 
1251 	task_io_accounting_init(&p->ioac);
1252 	acct_clear_integrals(p);
1253 
1254 	posix_cpu_timers_init(p);
1255 
1256 	do_posix_clock_monotonic_gettime(&p->start_time);
1257 	p->real_start_time = p->start_time;
1258 	monotonic_to_bootbased(&p->real_start_time);
1259 	p->io_context = NULL;
1260 	p->audit_context = NULL;
1261 	if (clone_flags & CLONE_THREAD)
1262 		threadgroup_change_begin(current);
1263 	cgroup_fork(p);
1264 #ifdef CONFIG_NUMA
1265 	p->mempolicy = mpol_dup(p->mempolicy);
1266 	if (IS_ERR(p->mempolicy)) {
1267 		retval = PTR_ERR(p->mempolicy);
1268 		p->mempolicy = NULL;
1269 		goto bad_fork_cleanup_cgroup;
1270 	}
1271 	mpol_fix_fork_child_flag(p);
1272 #endif
1273 #ifdef CONFIG_CPUSETS
1274 	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1275 	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1276 	seqcount_init(&p->mems_allowed_seq);
1277 #endif
1278 #ifdef CONFIG_TRACE_IRQFLAGS
1279 	p->irq_events = 0;
1280 	p->hardirqs_enabled = 0;
1281 	p->hardirq_enable_ip = 0;
1282 	p->hardirq_enable_event = 0;
1283 	p->hardirq_disable_ip = _THIS_IP_;
1284 	p->hardirq_disable_event = 0;
1285 	p->softirqs_enabled = 1;
1286 	p->softirq_enable_ip = _THIS_IP_;
1287 	p->softirq_enable_event = 0;
1288 	p->softirq_disable_ip = 0;
1289 	p->softirq_disable_event = 0;
1290 	p->hardirq_context = 0;
1291 	p->softirq_context = 0;
1292 #endif
1293 #ifdef CONFIG_LOCKDEP
1294 	p->lockdep_depth = 0; /* no locks held yet */
1295 	p->curr_chain_key = 0;
1296 	p->lockdep_recursion = 0;
1297 #endif
1298 
1299 #ifdef CONFIG_DEBUG_MUTEXES
1300 	p->blocked_on = NULL; /* not blocked yet */
1301 #endif
1302 #ifdef CONFIG_MEMCG
1303 	p->memcg_batch.do_batch = 0;
1304 	p->memcg_batch.memcg = NULL;
1305 #endif
1306 
1307 	/* Perform scheduler related setup. Assign this task to a CPU. */
1308 	sched_fork(p);
1309 
1310 	retval = perf_event_init_task(p);
1311 	if (retval)
1312 		goto bad_fork_cleanup_policy;
1313 	retval = audit_alloc(p);
1314 	if (retval)
1315 		goto bad_fork_cleanup_policy;
1316 	/* copy all the process information */
1317 	retval = copy_semundo(clone_flags, p);
1318 	if (retval)
1319 		goto bad_fork_cleanup_audit;
1320 	retval = copy_files(clone_flags, p);
1321 	if (retval)
1322 		goto bad_fork_cleanup_semundo;
1323 	retval = copy_fs(clone_flags, p);
1324 	if (retval)
1325 		goto bad_fork_cleanup_files;
1326 	retval = copy_sighand(clone_flags, p);
1327 	if (retval)
1328 		goto bad_fork_cleanup_fs;
1329 	retval = copy_signal(clone_flags, p);
1330 	if (retval)
1331 		goto bad_fork_cleanup_sighand;
1332 	retval = copy_mm(clone_flags, p);
1333 	if (retval)
1334 		goto bad_fork_cleanup_signal;
1335 	retval = copy_namespaces(clone_flags, p);
1336 	if (retval)
1337 		goto bad_fork_cleanup_mm;
1338 	retval = copy_io(clone_flags, p);
1339 	if (retval)
1340 		goto bad_fork_cleanup_namespaces;
1341 	retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1342 	if (retval)
1343 		goto bad_fork_cleanup_io;
1344 
1345 	if (pid != &init_struct_pid) {
1346 		retval = -ENOMEM;
1347 		pid = alloc_pid(p->nsproxy->pid_ns);
1348 		if (!pid)
1349 			goto bad_fork_cleanup_io;
1350 	}
1351 
1352 	p->pid = pid_nr(pid);
1353 	p->tgid = p->pid;
1354 	if (clone_flags & CLONE_THREAD)
1355 		p->tgid = current->tgid;
1356 
1357 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1358 	/*
1359 	 * Clear TID on mm_release()?
1360 	 */
1361 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1362 #ifdef CONFIG_BLOCK
1363 	p->plug = NULL;
1364 #endif
1365 #ifdef CONFIG_FUTEX
1366 	p->robust_list = NULL;
1367 #ifdef CONFIG_COMPAT
1368 	p->compat_robust_list = NULL;
1369 #endif
1370 	INIT_LIST_HEAD(&p->pi_state_list);
1371 	p->pi_state_cache = NULL;
1372 #endif
1373 	uprobe_copy_process(p);
1374 	/*
1375 	 * sigaltstack should be cleared when sharing the same VM
1376 	 */
1377 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1378 		p->sas_ss_sp = p->sas_ss_size = 0;
1379 
1380 	/*
1381 	 * Syscall tracing and stepping should be turned off in the
1382 	 * child regardless of CLONE_PTRACE.
1383 	 */
1384 	user_disable_single_step(p);
1385 	clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1386 #ifdef TIF_SYSCALL_EMU
1387 	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1388 #endif
1389 	clear_all_latency_tracing(p);
1390 
1391 	/* ok, now we should be set up.. */
1392 	if (clone_flags & CLONE_THREAD)
1393 		p->exit_signal = -1;
1394 	else if (clone_flags & CLONE_PARENT)
1395 		p->exit_signal = current->group_leader->exit_signal;
1396 	else
1397 		p->exit_signal = (clone_flags & CSIGNAL);
1398 
1399 	p->pdeath_signal = 0;
1400 	p->exit_state = 0;
1401 
1402 	p->nr_dirtied = 0;
1403 	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1404 	p->dirty_paused_when = 0;
1405 
1406 	/*
1407 	 * Ok, make it visible to the rest of the system.
1408 	 * We dont wake it up yet.
1409 	 */
1410 	p->group_leader = p;
1411 	INIT_LIST_HEAD(&p->thread_group);
1412 	p->task_works = NULL;
1413 
1414 	/* Now that the task is set up, run cgroup callbacks if
1415 	 * necessary. We need to run them before the task is visible
1416 	 * on the tasklist. */
1417 	cgroup_fork_callbacks(p);
1418 	cgroup_callbacks_done = 1;
1419 
1420 	/* Need tasklist lock for parent etc handling! */
1421 	write_lock_irq(&tasklist_lock);
1422 
1423 	/* CLONE_PARENT re-uses the old parent */
1424 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1425 		p->real_parent = current->real_parent;
1426 		p->parent_exec_id = current->parent_exec_id;
1427 	} else {
1428 		p->real_parent = current;
1429 		p->parent_exec_id = current->self_exec_id;
1430 	}
1431 
1432 	spin_lock(&current->sighand->siglock);
1433 
1434 	/*
1435 	 * Process group and session signals need to be delivered to just the
1436 	 * parent before the fork or both the parent and the child after the
1437 	 * fork. Restart if a signal comes in before we add the new process to
1438 	 * it's process group.
1439 	 * A fatal signal pending means that current will exit, so the new
1440 	 * thread can't slip out of an OOM kill (or normal SIGKILL).
1441 	*/
1442 	recalc_sigpending();
1443 	if (signal_pending(current)) {
1444 		spin_unlock(&current->sighand->siglock);
1445 		write_unlock_irq(&tasklist_lock);
1446 		retval = -ERESTARTNOINTR;
1447 		goto bad_fork_free_pid;
1448 	}
1449 
1450 	if (clone_flags & CLONE_THREAD) {
1451 		current->signal->nr_threads++;
1452 		atomic_inc(&current->signal->live);
1453 		atomic_inc(&current->signal->sigcnt);
1454 		p->group_leader = current->group_leader;
1455 		list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1456 	}
1457 
1458 	if (likely(p->pid)) {
1459 		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1460 
1461 		if (thread_group_leader(p)) {
1462 			if (is_child_reaper(pid))
1463 				p->nsproxy->pid_ns->child_reaper = p;
1464 
1465 			p->signal->leader_pid = pid;
1466 			p->signal->tty = tty_kref_get(current->signal->tty);
1467 			attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1468 			attach_pid(p, PIDTYPE_SID, task_session(current));
1469 			list_add_tail(&p->sibling, &p->real_parent->children);
1470 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
1471 			__this_cpu_inc(process_counts);
1472 		}
1473 		attach_pid(p, PIDTYPE_PID, pid);
1474 		nr_threads++;
1475 	}
1476 
1477 	total_forks++;
1478 	spin_unlock(&current->sighand->siglock);
1479 	write_unlock_irq(&tasklist_lock);
1480 	proc_fork_connector(p);
1481 	cgroup_post_fork(p);
1482 	if (clone_flags & CLONE_THREAD)
1483 		threadgroup_change_end(current);
1484 	perf_event_fork(p);
1485 
1486 	trace_task_newtask(p, clone_flags);
1487 
1488 	return p;
1489 
1490 bad_fork_free_pid:
1491 	if (pid != &init_struct_pid)
1492 		free_pid(pid);
1493 bad_fork_cleanup_io:
1494 	if (p->io_context)
1495 		exit_io_context(p);
1496 bad_fork_cleanup_namespaces:
1497 	if (unlikely(clone_flags & CLONE_NEWPID))
1498 		pid_ns_release_proc(p->nsproxy->pid_ns);
1499 	exit_task_namespaces(p);
1500 bad_fork_cleanup_mm:
1501 	if (p->mm)
1502 		mmput(p->mm);
1503 bad_fork_cleanup_signal:
1504 	if (!(clone_flags & CLONE_THREAD))
1505 		free_signal_struct(p->signal);
1506 bad_fork_cleanup_sighand:
1507 	__cleanup_sighand(p->sighand);
1508 bad_fork_cleanup_fs:
1509 	exit_fs(p); /* blocking */
1510 bad_fork_cleanup_files:
1511 	exit_files(p); /* blocking */
1512 bad_fork_cleanup_semundo:
1513 	exit_sem(p);
1514 bad_fork_cleanup_audit:
1515 	audit_free(p);
1516 bad_fork_cleanup_policy:
1517 	perf_event_free_task(p);
1518 #ifdef CONFIG_NUMA
1519 	mpol_put(p->mempolicy);
1520 bad_fork_cleanup_cgroup:
1521 #endif
1522 	if (clone_flags & CLONE_THREAD)
1523 		threadgroup_change_end(current);
1524 	cgroup_exit(p, cgroup_callbacks_done);
1525 	delayacct_tsk_free(p);
1526 	module_put(task_thread_info(p)->exec_domain->module);
1527 bad_fork_cleanup_count:
1528 	atomic_dec(&p->cred->user->processes);
1529 	exit_creds(p);
1530 bad_fork_free:
1531 	free_task(p);
1532 fork_out:
1533 	return ERR_PTR(retval);
1534 }
1535 
1536 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1537 {
1538 	memset(regs, 0, sizeof(struct pt_regs));
1539 	return regs;
1540 }
1541 
1542 static inline void init_idle_pids(struct pid_link *links)
1543 {
1544 	enum pid_type type;
1545 
1546 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1547 		INIT_HLIST_NODE(&links[type].node); /* not really needed */
1548 		links[type].pid = &init_struct_pid;
1549 	}
1550 }
1551 
1552 struct task_struct * __cpuinit fork_idle(int cpu)
1553 {
1554 	struct task_struct *task;
1555 	struct pt_regs regs;
1556 
1557 	task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1558 			    &init_struct_pid, 0);
1559 	if (!IS_ERR(task)) {
1560 		init_idle_pids(task->pids);
1561 		init_idle(task, cpu);
1562 	}
1563 
1564 	return task;
1565 }
1566 
1567 /*
1568  *  Ok, this is the main fork-routine.
1569  *
1570  * It copies the process, and if successful kick-starts
1571  * it and waits for it to finish using the VM if required.
1572  */
1573 long do_fork(unsigned long clone_flags,
1574 	      unsigned long stack_start,
1575 	      struct pt_regs *regs,
1576 	      unsigned long stack_size,
1577 	      int __user *parent_tidptr,
1578 	      int __user *child_tidptr)
1579 {
1580 	struct task_struct *p;
1581 	int trace = 0;
1582 	long nr;
1583 
1584 	/*
1585 	 * Do some preliminary argument and permissions checking before we
1586 	 * actually start allocating stuff
1587 	 */
1588 	if (clone_flags & CLONE_NEWUSER) {
1589 		if (clone_flags & CLONE_THREAD)
1590 			return -EINVAL;
1591 		/* hopefully this check will go away when userns support is
1592 		 * complete
1593 		 */
1594 		if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1595 				!capable(CAP_SETGID))
1596 			return -EPERM;
1597 	}
1598 
1599 	/*
1600 	 * Determine whether and which event to report to ptracer.  When
1601 	 * called from kernel_thread or CLONE_UNTRACED is explicitly
1602 	 * requested, no event is reported; otherwise, report if the event
1603 	 * for the type of forking is enabled.
1604 	 */
1605 	if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1606 		if (clone_flags & CLONE_VFORK)
1607 			trace = PTRACE_EVENT_VFORK;
1608 		else if ((clone_flags & CSIGNAL) != SIGCHLD)
1609 			trace = PTRACE_EVENT_CLONE;
1610 		else
1611 			trace = PTRACE_EVENT_FORK;
1612 
1613 		if (likely(!ptrace_event_enabled(current, trace)))
1614 			trace = 0;
1615 	}
1616 
1617 	p = copy_process(clone_flags, stack_start, regs, stack_size,
1618 			 child_tidptr, NULL, trace);
1619 	/*
1620 	 * Do this prior waking up the new thread - the thread pointer
1621 	 * might get invalid after that point, if the thread exits quickly.
1622 	 */
1623 	if (!IS_ERR(p)) {
1624 		struct completion vfork;
1625 
1626 		trace_sched_process_fork(current, p);
1627 
1628 		nr = task_pid_vnr(p);
1629 
1630 		if (clone_flags & CLONE_PARENT_SETTID)
1631 			put_user(nr, parent_tidptr);
1632 
1633 		if (clone_flags & CLONE_VFORK) {
1634 			p->vfork_done = &vfork;
1635 			init_completion(&vfork);
1636 			get_task_struct(p);
1637 		}
1638 
1639 		wake_up_new_task(p);
1640 
1641 		/* forking complete and child started to run, tell ptracer */
1642 		if (unlikely(trace))
1643 			ptrace_event(trace, nr);
1644 
1645 		if (clone_flags & CLONE_VFORK) {
1646 			if (!wait_for_vfork_done(p, &vfork))
1647 				ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1648 		}
1649 	} else {
1650 		nr = PTR_ERR(p);
1651 	}
1652 	return nr;
1653 }
1654 
1655 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1656 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1657 #endif
1658 
1659 static void sighand_ctor(void *data)
1660 {
1661 	struct sighand_struct *sighand = data;
1662 
1663 	spin_lock_init(&sighand->siglock);
1664 	init_waitqueue_head(&sighand->signalfd_wqh);
1665 }
1666 
1667 void __init proc_caches_init(void)
1668 {
1669 	sighand_cachep = kmem_cache_create("sighand_cache",
1670 			sizeof(struct sighand_struct), 0,
1671 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1672 			SLAB_NOTRACK, sighand_ctor);
1673 	signal_cachep = kmem_cache_create("signal_cache",
1674 			sizeof(struct signal_struct), 0,
1675 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1676 	files_cachep = kmem_cache_create("files_cache",
1677 			sizeof(struct files_struct), 0,
1678 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1679 	fs_cachep = kmem_cache_create("fs_cache",
1680 			sizeof(struct fs_struct), 0,
1681 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1682 	/*
1683 	 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1684 	 * whole struct cpumask for the OFFSTACK case. We could change
1685 	 * this to *only* allocate as much of it as required by the
1686 	 * maximum number of CPU's we can ever have.  The cpumask_allocation
1687 	 * is at the end of the structure, exactly for that reason.
1688 	 */
1689 	mm_cachep = kmem_cache_create("mm_struct",
1690 			sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1691 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1692 	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1693 	mmap_init();
1694 	nsproxy_cache_init();
1695 }
1696 
1697 /*
1698  * Check constraints on flags passed to the unshare system call.
1699  */
1700 static int check_unshare_flags(unsigned long unshare_flags)
1701 {
1702 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1703 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1704 				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1705 		return -EINVAL;
1706 	/*
1707 	 * Not implemented, but pretend it works if there is nothing to
1708 	 * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1709 	 * needs to unshare vm.
1710 	 */
1711 	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1712 		/* FIXME: get_task_mm() increments ->mm_users */
1713 		if (atomic_read(&current->mm->mm_users) > 1)
1714 			return -EINVAL;
1715 	}
1716 
1717 	return 0;
1718 }
1719 
1720 /*
1721  * Unshare the filesystem structure if it is being shared
1722  */
1723 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1724 {
1725 	struct fs_struct *fs = current->fs;
1726 
1727 	if (!(unshare_flags & CLONE_FS) || !fs)
1728 		return 0;
1729 
1730 	/* don't need lock here; in the worst case we'll do useless copy */
1731 	if (fs->users == 1)
1732 		return 0;
1733 
1734 	*new_fsp = copy_fs_struct(fs);
1735 	if (!*new_fsp)
1736 		return -ENOMEM;
1737 
1738 	return 0;
1739 }
1740 
1741 /*
1742  * Unshare file descriptor table if it is being shared
1743  */
1744 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1745 {
1746 	struct files_struct *fd = current->files;
1747 	int error = 0;
1748 
1749 	if ((unshare_flags & CLONE_FILES) &&
1750 	    (fd && atomic_read(&fd->count) > 1)) {
1751 		*new_fdp = dup_fd(fd, &error);
1752 		if (!*new_fdp)
1753 			return error;
1754 	}
1755 
1756 	return 0;
1757 }
1758 
1759 /*
1760  * unshare allows a process to 'unshare' part of the process
1761  * context which was originally shared using clone.  copy_*
1762  * functions used by do_fork() cannot be used here directly
1763  * because they modify an inactive task_struct that is being
1764  * constructed. Here we are modifying the current, active,
1765  * task_struct.
1766  */
1767 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1768 {
1769 	struct fs_struct *fs, *new_fs = NULL;
1770 	struct files_struct *fd, *new_fd = NULL;
1771 	struct nsproxy *new_nsproxy = NULL;
1772 	int do_sysvsem = 0;
1773 	int err;
1774 
1775 	err = check_unshare_flags(unshare_flags);
1776 	if (err)
1777 		goto bad_unshare_out;
1778 
1779 	/*
1780 	 * If unsharing namespace, must also unshare filesystem information.
1781 	 */
1782 	if (unshare_flags & CLONE_NEWNS)
1783 		unshare_flags |= CLONE_FS;
1784 	/*
1785 	 * CLONE_NEWIPC must also detach from the undolist: after switching
1786 	 * to a new ipc namespace, the semaphore arrays from the old
1787 	 * namespace are unreachable.
1788 	 */
1789 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1790 		do_sysvsem = 1;
1791 	err = unshare_fs(unshare_flags, &new_fs);
1792 	if (err)
1793 		goto bad_unshare_out;
1794 	err = unshare_fd(unshare_flags, &new_fd);
1795 	if (err)
1796 		goto bad_unshare_cleanup_fs;
1797 	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1798 	if (err)
1799 		goto bad_unshare_cleanup_fd;
1800 
1801 	if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1802 		if (do_sysvsem) {
1803 			/*
1804 			 * CLONE_SYSVSEM is equivalent to sys_exit().
1805 			 */
1806 			exit_sem(current);
1807 		}
1808 
1809 		if (new_nsproxy) {
1810 			switch_task_namespaces(current, new_nsproxy);
1811 			new_nsproxy = NULL;
1812 		}
1813 
1814 		task_lock(current);
1815 
1816 		if (new_fs) {
1817 			fs = current->fs;
1818 			spin_lock(&fs->lock);
1819 			current->fs = new_fs;
1820 			if (--fs->users)
1821 				new_fs = NULL;
1822 			else
1823 				new_fs = fs;
1824 			spin_unlock(&fs->lock);
1825 		}
1826 
1827 		if (new_fd) {
1828 			fd = current->files;
1829 			current->files = new_fd;
1830 			new_fd = fd;
1831 		}
1832 
1833 		task_unlock(current);
1834 	}
1835 
1836 	if (new_nsproxy)
1837 		put_nsproxy(new_nsproxy);
1838 
1839 bad_unshare_cleanup_fd:
1840 	if (new_fd)
1841 		put_files_struct(new_fd);
1842 
1843 bad_unshare_cleanup_fs:
1844 	if (new_fs)
1845 		free_fs_struct(new_fs);
1846 
1847 bad_unshare_out:
1848 	return err;
1849 }
1850 
1851 /*
1852  *	Helper to unshare the files of the current task.
1853  *	We don't want to expose copy_files internals to
1854  *	the exec layer of the kernel.
1855  */
1856 
1857 int unshare_files(struct files_struct **displaced)
1858 {
1859 	struct task_struct *task = current;
1860 	struct files_struct *copy = NULL;
1861 	int error;
1862 
1863 	error = unshare_fd(CLONE_FILES, &copy);
1864 	if (error || !copy) {
1865 		*displaced = NULL;
1866 		return error;
1867 	}
1868 	*displaced = task->files;
1869 	task_lock(task);
1870 	task->files = copy;
1871 	task_unlock(task);
1872 	return 0;
1873 }
1874